The Case for Cross-Border AI Incident Infrastructure

The Governance Gap in Frontier AI Systems

Countries around the world are becoming increasingly concerned about transnational harms resulting from the design, use and behaviours of artificial intelligence (AI). As a result, many countries voice aspirations to govern the development and deployment of general-purpose AI systems in a manner consistent with their values and strategic priorities.

Following in the footsteps of the United States and China, where the most capable (“frontier”) AI models and systems are developed, countries such as the United Kingdom, France, Canada, Japan, and India have committed billions in public funding to sovereign compute infrastructure and indigenous AI development.

Regions such as Africa and the European Union (EU) continue to collectively, and at the country level, articulate similar ambitions. Yet as these ambitions grow, so does the evidence base that the domestic institutions responsible for governing AI are still insufficient to address the rising harms produced by AI systems.

Why Cross-Border AI Incident Infrastructure is Needed

The damage produced by general-purpose AI systems does not respect borders, and existing mechanisms for sharing and responding to cross-border AI risks are inadequate for effective governance. In a tragic and recent example, in February 2026, eight people (including five children) were killed and 27 others were injured in Tumbler Ridge, Canada, at the hands of a shooter who had previously expressed their intent to kill during their interactions with ChatGPT, and had even used the chatbot to plan their attack.

Even though OpenAI staff acknowledged that this user’s interactions with ChatGPT were an indication of potential real-world violence and debated alerting local Canadian authorities, they ultimately decided not to contact law enforcement. This case effectively demonstrates that, without dedicated cross-border governance infrastructure, AI incidents and hazards are likely to continue causing significant societal harm. 

Understanding AI Risks and AI Hazards

“AI incidents” refer to events where the use or malfunction of AI systems leads to real-world harm to individuals, organizations, communities, property, or the environment.

Relatedly, “AI hazards” refer to risk-relevant findings that suggest that the development, use, or malfunction of an AI system could plausibly lead to incidents.

These span from implications on children’s safety and societal mental health to the malicious use of AI for cyberattacks and chemical, biological, radiological, and nuclear (CBRN) weapons development. Since AI models and systems built in one country are deployed worldwide, if left unaddressed, many such incidents and hazards risk escalating into large-scale emergencies, disasters, or crises.

Current Limitations in AI Incident Detection and Reporting

Present-day AI incident and hazard detection depends solely on voluntary disclosures from frontier AI companies, whistleblowers, social media self-reports, and journalistic coverage. Public incident databases, such as the Organisation for Economic Cooperation and Development’s (OECD) AI Incidents and Hazards Monitor, use these sources to record incidents and analyze trends across various harm categories.

Such databases show a significant increase in reported incidents over time, particularly those relating to cybercrime and flare-ups in mental health issues in users, with monthly reported incidents rising nineteen-fold over the past six years.

However, these databases only pick up and surface relevant patterns after serious harm has already occurred. This makes anticipatory governance extremely difficult. What the global AI governance community needs is for lessons from materialized AI harms to be collected and distributed widely, so that all jurisdictions can collectively learn from them.

Building Effective Cross-Border AI Incident Infrastructure

To ensure that countries are equipped to effectively detect, mitigate and respond to AI incidents, it is crucial that institutions tasked with national and international coordination on governing AI establish the mechanisms needed for doing so.

In our latest report, we make the case for cross-border AI incident infrastructure: institutional, technical, and legal arrangements that would allow national and international authorities to detect, share, and act on AI incidents and hazards together across borders.

Six Core Components of Cross- Border AI Incident Infrastructure

The six architectural components required for cross-border AI Incident infrastructure, along with concrete actions needed to secure them, are outlined below:

1. Internationally interoperable protocols for incident documentation and reporting. Governments and industry actors should adopt incident documentation structures and protocols that are universally compatible with one another, so that incident data collected in one jurisdiction becomes evidence that other jurisdictions can use.
2. Mandatory incident documentation, disclosure, and preparedness obligations on frontier AI companies. Governments should enact statutory incident reporting obligations on frontier AI companies and require them to publish policies for preventing, preparing for and responding to serious AI incidents.
3. Domestic incident data access and institutional coordination. Governments should grant national AI Safety Institutes, or functionally equivalent bodies, statutory access to incident data held by AI deployers and sectoral regulators, paired with formal information-sharing arrangements between such bodies and regulatory authorities.
4. Adequate international incident analysis capacity and expertise. International standards development organizations should develop a standardized methodology for AI incident root-cause and supply-chain analysis, with multilateral institutions such as the OECD and the United Nations (UN) advocating for its adoption and national bodies investing in the technical capacities needed to apply it.
5. Cross-border incident-sharing channels and joint investigation mechanisms. National  AI governance bodies, cybersecurity agencies, and sector-specific regulators should be empowered to share incident intelligence with counterparts abroad, such as through the International Network for Advanced AI Measurement, Evaluation and Science or through a new international mechanism that could be proposed via the UN’s Global Dialogue on AI Governance.
6. Internationally distributed incident detection, analysis and response capacity. Multilateral institutions should help extend incident infrastructure to  jurisdictions who have inadequate domestic incident monitoring and management capacities through the establishment of bilateral or multilateral intelligence-sharing, capacity-building, and incident response coordination partnerships between countries that already possess relatively mature incident infrastructure and those who do not.

Why Policymakers Must Act on Cross-Border AI Incident Infrastructure

As the Tumbler Ridge tragedy painfully illustrates, the cost of relying on fragmented, voluntary AI governance efforts are already being measured in human lives. The borderless nature of frontier AI systems demands an equally borderless infrastructure to detect, share, and mitigate hazards before they escalate into global crises.

We urge policymakers, multilateral organizations, and industry leaders to read the full report and begin implementing these six architectural pillars immediately. We can no longer afford to wait for the next tragedy to force our hand; the time to build a cohesive, international AI incident response network is now.

For more insights, research, and expert analysis, follow The Future Society on LinkedIn.